Yagil Osem, Institute of Plant Sciences and Genetics in Agriculture, Faculty of Agricultural, Food and Environmental Quality Sciences, The Hebrew University of Jerusalem, Rehovot, Israel (e-mail Osem@agri.huji.ac.il).
1The interactive effect of grazing and small-scale variation in primary productivity on the diversity of an annual plant community was studied in a semiarid Mediterranean rangeland in Israel over 4 years. The response of the community to protection from sheep grazing by fenced exclosures was compared in four neighbouring topographic sites (south- and north-facing slopes, hilltop and wadi (dry stream) shoulders), differing in vegetation, physical characteristics and soil resources. The herbaceous annual vegetation was highly diverse, including 128 species. Average small-scale species richness of annuals ranged between 5 and 16 species within a 20 × 20 cm quadrat, and was strongly affected by year and site.
2Above-ground potential productivity at peak season (i.e. in fenced subplots) was typical of semiarid ecosystems (10–200 g m−2), except on wadi shoulders (up to 700 g m−2), where it reached the range of subhumid grassland ecosystems. Grazing increased richness in the high productivity site (i.e. wadi), but did not affect, or reduced, it in the low productivity sites (south- and north-facing slopes, hilltop). Under grazing, species richness was positively and linearly related to potential productivity along the whole range of productivity. Without grazing, this relationship was observed only at low productivity (< 200 g m−2).
3The effect of grazing along the productivity gradient on different components of richness was analysed. At low productivity, number of abundant, common and rare species all tended to increase with productivity, both with and without grazing. Rare species increased three times compared with common and abundant species. At high productivity, only rare species continued to increase with productivity under grazing, while in the absence of grazing species number in the different abundance groups was not related to productivity.
4In this semiarid Mediterranean rangeland, diversity of the annual plant community is determined by the interaction between grazing and small-scale spatial and temporal variation in primary productivity, operating mainly on the less abundant species in the community.
The effect of grazing by large herbivores on the diversity of plant communities has been investigated in different terrestrial ecosystems (Milchunas & Lauenroth 1993; Proulx & Mazumder 1998). Grazing increased, reduced or lacked consistent effect on plant diversity (Huston 1994; Proulx & Mazumder 1998). These contrasting patterns of response have frequently been attributed to differences in grazing intensity, with greatest diversity expected at intermediate levels of grazing (Grime 1973; Connell 1978). More recent research, however, has shown that characteristics of the ecosystem that is subjected to grazing, such as primary productivity, evolutionary history and resulting vegetation physiognomy and plant life-forms, can interact with grazing in determining plant community structure and diversity (Whittaker 1977; Huston 1979, 1994; Milchunas et al. 1988; Milchunas & Lauenroth 1993; Noy-Meir 1995; Proulx & Mazumder 1998). Among these characteristics, primary productivity is particularly important as it determines both standing biomass and extent of grazing, while at the same time it modulates plant interactions and is linked to community structure (Waide et al. 1999; Grace & Jutila 1999). Furthermore, across plant communities, primary productivity is a comparable indicator of spatial and temporal variation in resource availability. Productivity–diversity relationships have been studied in varied ecosystems, with unimodal, positive and negative relationships frequently emerging (Grace & Jutila 1999; Waide et al. 1999; Gross et al. 2000). Usually, diversity is low in environments with very low availability of resources (i.e. where few species can survive) and increases with increasing resource availability. In contrast, diversity tends to decline in high productivity environments, due to competitive exclusion by favoured species that became abundant under these conditions (Grime 1973, 1979; Keddy 1990; Huston 1994). The significance of competition for generating productivity–diversity relationships in low-productivity systems is, however, not yet clear (Goldberg & Novoplansky 1997).
The impact of grazing on diversity differs along gradients of primary productivity (Milchunas et al. 1988; Louda et al. 1990; Okansen 1990), but there is no general consensus about the processes involved in this interaction. Competitive relationships among plants depend on resource availability. In environments with high resource availability and productivity, plants will be more likely to compete for canopy resources (light), while in less productive environments plant growth and diversity will be limited by soil resources (water, minerals) (Tilman 1982, 1988; Kadmon 1995). Shifts in the relative availability of canopy vs. soil resources might modulate interspecific competition and therefore the outcome of grazing effects on community structure. In his ‘dynamic equilibrium model of species diversity’Huston (1979) predicted that grazing can change diversity in opposite ways in resource-poor vs. resource-rich ecosystems, and this was supported by the meta-analysis performed over a wide variety of environments by Proulx & Mazumder (1998) (their grazing reversal hypothesis). Furthermore, based on a world-wide data-set analysis, Milchunas & Lauenroth (1993) proposed that alteration in grassland structure and diversity due to grazing was primarily a function of productivity and evolutionary grazing history of each particular grassland ecosystem. These factors determine species composition and the prevailing life-forms and morphological traits that characterize the plant community. This, in turn, dictates the responses of individual species to grazing.
The interaction between grazing and primary productivity on species diversity has been mostly studied at the ecosystem level, in perennial grasslands (Milchunas & Lauenroth 1993). In contrast to perennial grasslands, annual grasslands lack temporal continuity in their competitive interactions. Furthermore, in perennial grasslands the window of relaxed competitive interaction after disturbance is mainly a consequence of plant regrowth capabilities, while annual grasslands depend more on seed-bank dynamics and seedling establishment (Briske & Noy-Meir 1998). We asked if the relationships found between diversity, productivity and grazing in perennial grasslands are operative also in annual grasslands, particularly in low productivity, semiarid regions. The herbaceous communities in Mediterranean semiarid rangelands offer an excellent opportunity to study this question, as they have been used for grazing since historical times and are dominated by annual species, many of them tolerant to grazing (Noy-Meir & Seligman 1979; Le Houerou 1993; Shmida & Ellner 1983; Perevolotsky & Seligman 1998). Annuals are widespread in Mediterranean rangelands and well adapted to semiarid ecosystems (Shmida & Burgess 1988), in which primary productivity is usually limited by seasonality in soil resources, mainly water and nitrogen (Seligman & van Keulen 1989). Furthermore, the typical heterogeneity of annual rainfall and habitat topography in semiarid regions often results in a highly patchy distribution of these limiting resources (Noy-Meir 1973). This heterogeneity, together with the dynamic nature of annual plant populations that regenerate every year from the available seed-bank, can cause wide spatial and temporal variation in primary productivity.
In this study, we propose that the Milchunas et al. (1988) general model developed for geographically separated ecosystems (i.e. large scale) differing in primary productivity, also applies to local fine scales. We hypothesize that in Mediterranean semiarid regions, in which annual plants are the main primary biomass producers, changes in community diversity in response to grazing will depend on small-scale variation of productivity in time and space. More specifically, we asked: (i) Does the productivity-richness relationship vary between topographic sites and between years that differ in productivity? (ii) Does the extent and direction of change in richness caused by grazing depend on small-scale temporal and spatial variation in productivity? (iii) Which components of species richness contribute to such changes in diversity? These questions were studied in a long-term experiment in a semiarid rangeland in Israel, by comparing the responses of annual plant communities to protection from grazing in neighbouring topographic sites that differ in primary productivity.
Materials and methods
the study area
The research was conducted at the LTER Lehavim Research Station (Bedouin Demonstration Farm), located in the Goral Hills, near Beer-Sheva in the Northern Negev, and within the Irano-Turanian phytogeographical region. Due to steep climatic gradients, the Irano-Turanian region is present as a narrow strip between the Mediterranean and the Saharo-Arabian phytogeographical regions (Zohary 1973). The vegetation physiognomy in the study area is a sparse shrub-land in which Sarcopoterium spinosus, Corydothymus capitatus and Thymelea hirsuta are the dominant shrubs. Annual species represent 56% of the regional flora (Danin & Orshan 1990). The herbaceous vegetation appears shortly after the first rains and persists for 2–5 months, depending on the amount and distribution of rainfall. This vegetation is very diverse, mostly composed of annual species and has been used for livestock grazing (mainly sheep and goats) since pre-historical times (5000–8000 years) (Noy-Meir & Seligman 1979). Thus, the vegetation is mostly composed of grazing tolerant species, with less tolerant ones probably having become extinct (Perevolotsky & Seligman 1998).
Climate at the Lehavim Station is Mediterranean semiarid, with winter rains concentrated during December to March, and average annual precipitation of 305 mm (1951–95: max = 540 mm, min = 78 mm) (Baram 1996). Annual precipitation in the years 1996, 1997, 1998 and 1999 was 223, 247, 235 and 40 mm, respectively (1999 was the driest year on record). Average daily temperature ranges from 10 °C in the winter to 25 °C in the summer (Baram 1996). Soils are loamy textured, of a brown desert skeletal type developed on Eocene limestone, dolomite and chalk (Ravikovich 1981). Altitude is 350–500 m a.s.l. The area was grazed every year by a flock of about 600 Awassi sheep, starting in late January, after rainfall onset and plant establishment, continuing until May (green pasture), and again from August to December (dry pasture). A shepherd directed the flock continuously over the range, to ensure foraging over the entire area and maximal consumption of the available vegetation. Thus, grazing intensity varied according to the availability of above-ground biomass, with residual biomass of 25% and 50% in the high and low productive sites, respectively.
experiment design and sampling
This work is part of a long-term study on rangeland management at the Lehavim Station. The experiment was performed in a heavily grazed area near the sheep pen, in four neighbouring sites with different topographies: wadi shoulders (narrow flat areas beside dry streams), hilltop, and south- and north-facing slopes (south-facing slope, azimuth 210°, incline 12.5°; north-facing slope, azimuth 27°, incline 15°). Four permanent exclosures (10 × 10 m) to prevent sheep grazing were established in each topographic site in 1993. Although the sites are adjacent (370 m between the two most distant plots), they differ in characteristics such as solar radiation load, temperature, soil properties and water balance, which are determined by aspect and inclination. In particular, soil at the wadi site is deeper, has better water-holding capacity and gets additional water run-off from the slopes during rainfalls than the shallow soil, with low organic matter found in the other topographic sites. These differences in abiotic conditions are reflected in the density and composition of the herbaceous and woody vegetation in the different sites. In addition, shrubs are absent in the wadi shoulders, largely due to uprooting to increase grazing area and past use as fuel.
The study area is characterized by patchy distribution of rocks and shrubs (except the wadi), with open patches among them. Data on above-ground biomass, species richness and plant density, were gathered in each site at peak season (spring, April) during 4 years, between 1996 and 1999. Random samples of the herbaceous vegetation (20 × 20 cm quadrats) were clipped at soil level in the open patches. This quadrat size is commonly used to study Mediterranean herbaceous communities (Montalvo et al. 1993) and is appropriate to the small-scale range at which plant interactions occur in them. In each topographic site, the samples (quadrats) were taken from four plots (each exclosure and its surrounding area are considered a plot ‘nested’ within the site, in a pseudo replicated design). In each plot, five samples were clipped inside the exclosure (ungrazed subplot) and five outside it (grazed subplot). Plants were identified and counted to determine plant density, and their above-ground dry biomass was weighed separately by species, after 48 h oven drying (75 °C).
parameters and definitions
Annual species comprise more than 90% of the herbaceous biomass and peak-standing crop in each year is therefore a good indicator of annual productivity and thus resource availability in this rangeland. In this system the growing season is short and biomass loss due to litter prior to harvesting is very small. Therefore, the average biomass within an exclosure was taken as the actual primary productivity of the ungrazed subplot and as the potential primary productivity of the adjacent grazed subplot. Plant density was the average number of plants quadrat−1. Small-scale species diversity of the annual plant community was evaluated by species richness (average number of species quadrat−1) and by the Shannon-Wiener index of diversity (Magurran 1988). To assess the relative rareness or commonness of the species contributing to diversity along the productivity gradient, species were divided into three groups according to their relative abundance in the vegetation of the study area: abundant (more than 5% of total density), common (1–5%) and rare (less than 1%) species.
Repeated measures anovas were applied to study the effects of year (1996, 1997, 1998, 1999), site (wadi shoulders, hilltop, and south- and north-facing slopes) and fencing (grazed, ungrazed) in fully factorial analyses. Sampled plots were nested within site. Residuals of above-ground biomass, plant density and species richness lacked normality (Shapiro-Wilk W-test) and data were therefore ranked (ties averaged) and an anova applied to the ranked data. The analyses were therefore non-parametric (Conover & Iman 1981). Regression analyses were performed to study the relationships between productivity, plant density and species richness. The statistical package JMP (SAS Institute 1995) was used.
the annual vegetation
The herbaceous annual vegetation was highly diverse, with 128 species identified during the 4 years of sampling in the research area. Of these, 15 species contributed, on average, 85% of the plant density and of the above-ground biomass of annuals. Grasses represented about 46% of the total density and 51% of the above-ground biomass, and legumes 9% and 21%, respectively. The rest was contributed by a wide variety of annual forbs.
small-scale variation in productivity and plant density
Biomass of the protected subplots (i.e. productivity) varied greatly between years (F3,276 = 152, P < 0.0001) and topographic sites (F3,12 = 10.3, P < 0.0012), ranging from 10 g m−2 (hilltop 1999) to 494 g m−2 (wadi 1998) (Fig. 1). The two-way anova model applied to data from these ungrazed subplots explained 72% of the temporal and spatial variance in productivity, with year accounting for 64% and site for 24% of the explained variance. The interaction of year × site was small (3%; F9,276 = 2.68, P = 0.0053) and the effect of year was stronger in the wadi than in the other sites. Productivity in the wadi was consistently greater (two to five times) than in the other sites (P < 0.05), but no significant differences were found among the low productivity sites (hilltop, and south- and north-facing slopes). In each site, productivity in 1997 and 1998 was four to ten times larger than in 1999 (the driest year) (P < 0.05).
Above-ground biomass remaining at peak season in the fenced (ungrazed) subplots was up to four times higher than in the grazed subplots (F1,564 = 174, P < 0.0001) (Fig. 1). The three-way anova model explained 70% of variance in all grazed and ungrazed subplots, with year, site and fencing accounting for 61%, 19% and 14% of the explained variance, respectively. The effect of fencing on biomass increased with productivity and depended on both year (year × fencing: F3,564 = 4.6, P = 0.0035) and site (site × fencing: F3,564 = 3.39, P < 0.018), but the site × year × fencing interaction was not significant (P > 0.05). In the grazed subplots, above-ground biomass at peak season was similar in all three less productive sites, but was 16–67% lower than in the wadi (P < 0.05).
Plant density varied significantly among years (F3,564 = 54.1, P < 0.0001) and topographic sites (F3,12 = 8.05, P = 0.003), ranging between 24 and 132 plants per quadrat (Fig. 2). Plant density tended to be lowest in the south-facing slope, in both the grazed and ungrazed subplots (P < 0.05), with differences between the other sites small, and generally not significant, in spite of the much higher biomass in the wadi (Fig. 1). Protection from grazing usually increased plant density in the less productive sites (F1,564 = 12.0, P = 0.0006), while in the wadi negative or zero response occurred in some years (fencing × year: F3,564 = 7.35, P = 0.0001). The three-way anova model explained 38% of variance in plant density, with year, site and fencing accounting for 46%, 22% and 3.4% of the explained variance, respectively.
small-scale variation in species richness
Species richness of annuals ranged between five and 16 species per sample and was strongly affected by year (F3,564 = 80.7, P < 0.0001) and site (F3,12 = 11.5, P = 0.0007) (Fig. 3). The three-way anova model explained 44% of the variance, with year and site accounting for 54% and 20% of the explained variance, respectively. Richness was lowest in 1999, the driest year, and highest in 1997 and 1998, the years with highest productivity (P < 0.05). Effects of protection from grazing were more complex, with a large fencing–site interaction (F3,564 = 15.0, P < 0.0001), that accounted for 10% of the explained variance. Fencing reduced species richness in the wadi (F1,141 = 14.8, P = 0.0002), but increased it at all other sites, significantly so in the north-facing slope (F1,141 = 5.07, P = 0.026) and in the hilltop (F1,141 = 5.39, P = 0.022). Thus, protection from grazing had opposite effects on species richness in the wadi, the most productive site, compared with the less productive sites. Species richness of the grazed subplots was always highest in the wadi (P < 0.05), whereas in ungrazed subplots differences among sites were very small or absent. In 1999 (the extremely dry year) the positive effect of protection on species richness in the hilltop and north-facing slope disappeared, while its negative effect remained in the wadi site. We found similar trends, but with smaller differences between grazed and ungrazed subplots, in comparisons based on the Shannon-Wiener index of diversity (results not shown).
The possibility that the reversal in the grazing effect on richness at high productivity was due to greater intensity of grazing in the high productive site was analysed by plotting the differences in richness between grazed and protected subplots as a function of grazing intensity (i.e. percentage of above-ground biomass consumed). We found no significant linear correlation between grazing intensity and the changes in richness across the whole productivity range. Grazing was intense (60–80% biomass consumed) in both high and low productivity plots. Within this range, richness was significantly increased (2.69 ± 0.68, P < 0.05) by fencing in plots at the less productive sites, but the effect of exclusion was reversed in the high productive plots of the Wadi (−3.63 ± 0.84, P < 0.05).
relationships between productivity, plant density and species richness
The possibility that variation in the diversity of the annual plant community was due to small-scale spatial (plot) and temporal (year) differences in productivity and plant density was studied by regression analysis, based on plot data from the different topographic sites and years. Species richness and Shannon-Wiener indexes showed similar patterns of change and only richness data are therefore presented. Species richness in the grazed subplots was positively and linearly related to their potential productivity along the whole range of productivity over all years (Fig. 4). In the ungrazed subplots this linear relationship was observed only when productivity was below 200 g m−2 (hilltop, and south- and north-facing slopes). When productivity was above 200 g m−2 (mainly wadi shoulders), richness in ungrazed subplots was not related to differences in productivity due to plot and year, fluctuating widely between five and 16 species per quadrat (average 8.70 ± 0.29). In the grazed subplots, in contrast, the relationship remained significant at high productivity (R2 = 0.47, n = 13, P = 0.01). When grazed and ungrazed subplots were compared within the low productivity plots, productivity-richness relationships were weaker (R2 = 0.26, n = 51, P < 0.0001) and its slope smaller in the grazed subplots (P < 0.05).
Species richness was also related to plant density, showing trends similar to those found with productivity. Thus, in the grazed subplots, species richness increased along the whole range of plant densities (sites and years together) (R2 = 0.34, n = 64, P < 0.0001), but only below a productivity of 200 g m−2 in ungrazed subplots (R2 = 0.54, n = 51, P < 0.0001).
This similarity in the relationships between richness vs. productivity and plant density was partially due to the fact that productivity and plant density were related. Thus, in ungrazed subplots plant density increased with productivity (R2 = 0.26, n = 51, P < 0.0001) in the low productivity sites (hilltop, south- and north-facing slopes, wadi in dry 1999), but no relationship was found in the ungrazed high productivity subplots. In contrast, plant density in the grazed subplots increased along the whole range of potential productivity (R2 = 0.14, n = 64, P < 0.002).
components of species richness
We analysed the change along the productivity gradient in the number of abundant, common and rare species (Fig. 5). In the low productivity plots, the number of abundant and rare species increased significantly with productivity in both the grazed and ungrazed subplots, while common species increased only in the ungrazed subplots (Fig. 5; Table 1). The increase in abundant and rare species was about twice in the ungrazed subplots vs. grazed subplots, but the differences were not significant. On the other hand, increasing productivity had about a three times greater effect on rare as on common or abundant species, in both the grazed and ungrazed subplots. This difference was significant only in the ungrazed subplots (P < 0.05). In the high productivity plots, the change in the number of species in the different abundance groups was grazing dependent. Under grazing, rare species continued to increase with productivity at the same rate as in the low productivity plots, but the number of common and abundant species was no longer related to productivity (Table 1). In the absence of grazing, the number of rare, common and abundant species was not related to productivity. Thus, the rare species are the main contributors to the changes in species richness observed under different productivity levels and grazing.
Table 1. Relationship between the number of abundant (> 5% of total abundance), common (1–5%) and rare (< 1%) species and potential productivity in grazed and ungrazed subplots with low (< 200 g m−2; n = 51) and high (> 200 g m−2; n = 13) productivity levels. b= regression coefficient. NS = not significant
Low productivity plots (< 200 g m−2)
High productivity plots (> 200 g m−2)
Primary productivity of the annual plant community in this Mediterranean semiarid rangeland varied strongly through time and space, with water availability as a main driving factor, as found in other semiarid regions (Noy-Meir 1973; Le Houerou 1993). The temporal and spatial variation in productivity was mainly due to interannual differences in precipitation (40–247 mm year−1), and to heterogeneity in water run-off/run-on and physical characteristics among the topographic sites. During the study period, the above-ground productivity at peak season in the subplots protected from grazing in the hilltop, and south- and north-facing slopes was 10–200 g m−2, a range typical of semiarid ecosystems. In contrast, productivity in the wadi shoulders reached up to 700 g m−2, within the range for subhumid grassland ecosystems (Milchunas et al. 1988), although productivity in drier years (< 200 g m−2) was more characteristic of semiarid ecosystems. Within this productivity gradient, grazing increased richness in the high productivity site (wadi shoulders), but did not affect or reduced it in the low productivity sites (Fig. 2). This reversal in richness as a function of productivity was predicted and therefore provides further support to the dynamic equilibrium model (Huston 1979), Milchunas et al.'s (1988) generalized model and the grazing reversal hypothesis (Proulx & Mazumder 1998). Although it can be argued that the different effects of grazing at high productivity could be attributed to their intrinsically higher grazing intensity, this intensity was not correlated with the extent of change in richness across our productivity gradient. Higher diversity in grazed vs. ungrazed grasslands has also been reported for Mediterranean regions with higher rainfall and greater productivity (Naveh & Whittaker 1979; Noy-Meir et al. 1989; Montalvo et al. 1993; Hadar et al. 2000; Sternberg et al. 2000), as well as in coastal meadows (Grace & Jutila 1999) and other grassland systems (Waser & Price 1981; Pucheta et al. 1998; Dupre & Diekman 2001).
The opposite effects of grazing on species richness observed along the productivity gradient can be interpreted as the outcome of the interaction between grazing, resource availability and plant competition for different limiting resources, as shown in the conceptual model in Fig. 6. A basic assumption is that, at the low productivity (< 200 g m−2), plant growth and diversity is limited by soil resources (mainly water and minerals), while at higher levels of productivity, leading to larger above-ground biomass, competition is mainly for canopy resources (Tilman 1982, 1988; Milchunas et al. 1988). Thus, in the low productivity range, in which the gradual increase in richness can be related to increasing availability of soil resources, richness was either unaffected or slightly reduced by grazing, most likely due to plant removal and trampling (Noy-Meir 1990). At higher levels of productivity (such as in the wadi), species richness continues to rise under grazing, probably in response to larger soil resources and to a parallel reduction of competition for light, due to removal by grazing of the palatable larger species. Without grazing, on the other hand, richness did not continue to increase in the higher productivity range and was generally lower than in the grazed subplots. This lack of increase could result from greater competition for canopy resources (i.e. light) and displacement of smaller, less competitive species (Grime 1979; Huston 1994). All together, these trends can be interpreted as a reversal in richness due to grazing along a productivity gradient, in which plant competition shifts from below-ground to above-ground resources. Despite the widely held assumption that grazing lowers competition intensity, few studies actually examined the impact of grazing on competitive interactions among plants (Goldberg & Barton 1992; Taylor et al. 1997), and none of them have considered annual plant communities along gradients of productivity.
Milchunas et al. (1988) suggested that the shift from competition for soil resources to competition for canopy resources that occurs with increasing productivity might explain the differences in the response to grazing found between semiarid and subhumid grasslands. Our data suggest that this shift also explains the interaction between grazing and small-scale variation in productivity that determines species richness in heavily grazed annual plant communities in semiarid Mediterranean rangelands possessing high spatial and temporal heterogeneity in resource availability. Our small-scale conceptual model corresponds to the large-scale model proposed by Milchunas et al. (1988) for semiarid to subhumid ecosystems of perennial grasslands with a long history of grazing. It is important to note that, in contrast to the Milchunas et al. generalized model, which compared distinct communities occurring along geographical gradients in productivity, we compared neighbouring sites whose annual plant communities were relatively similar before the establishment of the exclosures. Sorenson's index of similarity among the high (wadi) and low productivity sites (hilltop, and north- and south-facing slopes), as well as among the low productivity sites themselves, was 0.5–0.6 (Osem et al., unpublished results). This further supports our conclusion that the changes in richness due to protection from grazing were due mainly to differences in productivity of the topographic sites and not to initial floristic differences in their annual plant communities.
Contrary to expectations, richness did not decrease at high productivity levels in the absence of grazing and the richness–productivity relationship was not therefore unimodal across sites and years. However, unimodal relationships have been found in other semiarid plant communities with comparable ranges of productivity, including some in the Mediterranean region (Kutiel & Danin 1986; Puerto et al. 1990). In a comparable study in a transition zone between grassland and shrubland in Arizona in which annuals are dominant, the hump-shaped relationship was detected only when very diverse microhabitats (open sites, ant and rat mounds, and shrub fertility islands) were pooled together in the analysis (Guo & Berry 1998). In the Chihuahua Desert, sites protected from grazing and with higher productivity due to additional ‘run-on’ were found to be overwhelmingly dominated by a single species and exhibited lower diversity, compared with less productive sites (Ludwig 1986). These facts suggest that a unimodal curve emerges when wide ranges of the environmental productivity gradient are integrated (Rosenzweig 1995; Pausas 2001). The large variation in species richness observed in the ungrazed subplots in the wadi (the most productive site) rather than the expected decrease suggests that other processes are masking the richness–productivity relationship. Possible explanations for this large variation in richness are the relatively short time that has elapsed since exclosures were established in 1993 and variation in the rate of recovery of the annual vegetation of different plots after cessation of grazing. Analysis of soil samples from the wadi site showed that seed density of the larger and palatable grasses and legumes in the grazed subplots is very low, due to intensive grazing and limited dispersal of their seeds (Osem et al., unpublished data). This may have resulted in a patchy distribution of the larger species during the initial stages of recovery after fencing, with their distribution becoming more homogeneous with time. Thus, if rate of recovery differs among the high productivity plots, it is conceivable that in plots in which the seed density of the larger and palatable species is still relatively low, the competitive displacement of the smaller species will be less intense, and richness in these plots will be higher. Additionally, a persistent seed bank of the less competitive species may lead to their slower decline in the vegetation after fencing, thus maintaining a higher diversity. It can be argued that, in annual grasslands, variation in species richness after cessation of grazing is modulated by initial seed density constraints and by a balance between the seed-bank dynamics of the different species. These trends in the vegetation and in the seed-bank most probably affect the relationship between productivity and richness during recovery from grazing, with the hump-shaped relationship emerging after longer periods of protection from grazing. Local deficiency of seeds has been proposed as an important factor in generating small-scale vegetation patterns in calcareous grasslands (Zobel et al. 2000; Turnbull et al. 2000).
The changes in richness that were observed in different plots along the productivity gradient could be attributed to variation in plant density, as increasing plant density should increase the probability of finding additional species. Indeed, the density vs. productivity relationships were similar to those observed for richness vs. productivity. However, the richness vs. density relationships were weaker compared with the richness vs. productivity relationships. Furthermore, the large variation in richness observed in the high productivity subplots protected from grazing was not related to density, thus indicating the involvement of other processes.
Analysis of the components of species richness revealed a consistent increase in rare species across the range of productivity, whereas numbers of abundant and common species remained relatively constant. The effects of productivity and grazing on the small-scale temporal and spatial diversity of the annual plant community were therefore due to variation in the number of less abundant species. It can be argued that variation in the presence of species is limited by their availability at a regional scale (the regional species pool; Zobel 1997; Grace 2001). Indeed, the study area is surrounded by grazed land with similar vegetation, from which species with low tolerance to grazing disappeared long ago. Moreover, dispersal capability of most species in the region (except a few composites with wind dispersal) is quite limited (Ellner & Shmida 1981; Osem et al., unpublished data). Therefore, short-term changes in the annual vegetation after grazing exclusion are less likely to be due to colonization by propagules originating from the regional species pool, but rather to result from shifts in the relative proportion of species initially present in the local, small-scale species pool. This pool is maintained in the seed-bank of the grazed area and in natural refuges protecting from grazing, such as rock outcrops and shrubs. In the Spanish dehesa annual pastures, rare species (representing 47% of the total richness) are assumed to persist in the community as a result of their persistent seed-banks, which allow establishment and production of fresh seed input in rainy (i.e. highly productive) years (Peco et al. 1998). It is conceivable that the establishment and reproductive success of species from the local pool is controlled by abiotic (i.e. resource availability) and biotic (i.e. grazing, competition) factors that operate at the small spatial-scale, and whose relative importance changes along the productivity gradient. Within this context, it can be argued that less abundant species are more susceptible to availability of resources and grazing (Zobel et al. 2000). Indeed, the number of the abundant species along the productivity gradient was relatively small and constant. Thus, in this semiarid Mediterranean rangeland with winter rains, diversity of the annual plant community is mainly determined by the less abundant species, due to the interaction between grazing and small-scale temporal and spatial variation in primary productivity.
This research was supported by the International Arid Land Consortium (IALC), grants 96R-19 and 98R-27. We thank Dany Milchunas, Imanuel Noy-Meir and two reviewers for critical reading of the original manuscript, helpful comments and suggestions, and to Hillary Voet for statistical advice. Many thanks to Marcelo Sternberg, Rana German, Zohar Shaham, Ram Lisai and Rafi Yonathan for their help in the fieldwork.